Add Apple GPU backend for count_overlaps

[mwiewior]

Summary

• add optional apple-gpu feature using objc2-metal and bio.count_overlaps_backend = auto|cpu|apple_gpu
• implement Apple Metal rank-based count_overlaps backend with CPU fallback, projection pushdown, right-side column pruning, and GPU batch coalescing
• add count/projection/unit/integration coverage plus release benchmark timing breakdowns
• document OpenSpec change for the Apple GPU backend and benchmarking/dispatch behavior

Benchmarks

8-7 dataset from /tmp/polars-bio-bench/databio (ex-rna vs ex-anno), release mode, Apple M3 Max, 1 thread:

• CPU mean: 611.363 ms
• final auto/GPU mean after optimizations: 402.442 ms
• result: n_rows=1194285,total_overlaps=307298602

Timing breakdown after final optimization:

• left collect: ~51-55 ms
• CPU rank-index/backend build: ~282-303 ms
• GPU batch execution: ~30-32 ms

Verification

• cargo test -p datafusion-bio-function-ranges --features apple-gpu
• cargo test -p datafusion-bio-function-ranges --features apple-gpu test_count_overlaps_apple_gpu_matches_cpu_fixtures -- --ignored --nocapture
• cargo run -p datafusion-bio-function-ranges --release --features apple-gpu --example count_overlaps_bench -- --backend auto --threads 1 --repeats 5 --warmups 1 --timings --left /tmp/polars-bio-bench/databio/ex-rna --right /tmp/polars-bio-bench/databio/ex-anno